Object Detection Part 3: Faster R-CNN, Region Proposal Network and Intersection over Union

[@datamlistic]

Check out the whole object detection series here: kzbin.info/aero/PL8hTotro6aVG6prsY92ZNVBNPr1PkXgsP

[@tfun-ef1pm]

thank you so much, the explaination and the demonstrations are so much easier to understand than the paper

[@datamlistic]

Glad you enjoyed it! Also make sure to check the other videos in the object detection series. :)

[@manasamani286]

thanks for your explanation. is this type of model suitable for detecting car liscence plates ? ( for blurring them. )

[@datamlistic]

You're wecome! Yes, you can use it for that, but I think there are better and faster models out there right now. Check out the latest versions of YOLO for instance.

[@ranjan561]

I have 1D Data, I want to apply faster RCNN , any resources for the same ?

[@datamlistic]

Hi there. Unfortunately, I am not aware of any faster RCNN implementations that work specifically with 1D data. I can give you some hints on how you can implement a faster RCNN on your own for 1D data with 1D convolutions if you wish. Also, you can also try to use the plain faster RCNN model with 2D convolutions by reshaping your data to be something like (max_len, 1, 1) and artificially set the labeled bounding box y coordinates to 0, while on the x axis you have the boxes you wish to detect. In addition, you have to be careful on how you do the RoI/Max pooling because you have to make the algorithm return only one value on y for each x bin. I hope this makes sense. Please let me know if you have any other questions. :)

[@Bapll]

I was stuck for about an hour or so, looking at the Object classifier and Bounding Box Regressor, thinking that "2k" and "4k" meant 2000 and 4000. Funnily enough, I couldn't get it to make sense in my head. My god, I need to sleep or something...

[@datamlistic]

Haha, could happen to anyone. Take care of your sleep, mate! :)

[@alexis91459]

Very well explained as always. Idea for a video on a point I have trouble with: why infinite width bayesian deep networks are gaussian processes

[@datamlistic]

Thank you for your suggestion! I've added it to my list. Let me know if you have other subjects you would like to see. :)

[@alexis91459]

@@datamlistic Some ideas that interest me: -the different architectures in GNN -optimizations in ML (mixed precision, locality-sensitive hashing, etc) -More exotic architectures (Euclidean neural networks,...) Hoping that this will be useful for the future!

[@datamlistic]

@@alexis91459 Thank you so much for your feedback! I've also added those subjects on my list. :)

[@hussamcheema]

Its confusing.

[@datamlistic]

Could you elaborate what you've found confusing about this explanation?